HAL Id: hal-02499712
https://hal.archives-ouvertes.fr/hal-02499712
Submitted on 5 Mar 2020
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
with general scalar diffusive functions
Georges Chamoun, Mazen Saad, Raafat Talhouk
To cite this version:
Georges Chamoun, Mazen Saad, Raafat Talhouk. Finite volume scheme for isotropic Keller-Segel
model with general scalar diffusive functions. ESAIM: Proceedings and Surveys, 45, pp.128 - 137,
2014, �10.1051/proc/201445013�. �hal-02499712�
ESAIM: PROCEEDINGS AND SURVEYS,September 2014, Vol. 45, p. 128-137 J.-S. Dhersin, Editor
FINITE VOLUME SCHEME FOR ISOTROPIC KELLER-SEGEL MODEL WITH
GENERAL SCALAR DIFFUSIVE FUNCTIONS
∗Georges Chamoun
1,2, Saad Mazen
1and Talhouk Raafat
2Abstract. This paper is devoted to the numerical analysis of a modified Keller-Segel model consisting of diffusion and chemotaxis with volume filling effect. Firstly, a finite volume scheme is generalized to the case of a Keller-Segel model allowing heterogeneities and discontinuities in the diffusion coefficients. For that, we start with the derivation of the discrete problem and then we establish a convergence result of the discrete solution to a weak solution of the continuous model. Finally, numerical tests illustrate the behavior of the solutions of this generalized numerical scheme.
R´esum´e. Cet article est consacr´e `a l’analyse num´erique d’un mod`ele g´en´eralis´e de Keller-Segel mod´elisant l’interaction entre la densit´e des cellules et la concentration d’un chimio-attractant. La diffusion des cellules est mod´elis´ee par un op´erateur d´eg´en´er´e en ´evitant l’explosion de la densit´e cellu-laire autour du chimio-attractant grˆace au principe du ”volume filling effect”. D’abord, un sch´ema de volumes finis est g´en´eralis´e au cas de mod`eles d´eg´en´er´es de Keller-Segel avec des coefficients diffusifs h´et´erog`enes discontinus. Ensuite, on montre la convergence des solutions du probl`eme discret vers une solution faible du probl`eme continu. Enfin, des tests num´eriques illustrent le comportement des solutions de ce sch´ema num´erique g´en´eralis´e.
1. Introduction
From microscopic bacteria through to the largest mammals, the survival of many organisms is dependent on their ability to navigate, through the detection of internal and external signals. The ability to migrate in response to chemical gradients, chemotaxis, has attracted significant interest due to its critical role in a wide range of biological phenomena. Mathematical modeling of chemotaxis has developed into a large and diverse discipline, whose aspects include its mechanistic basis and the modeling of specific systems. The Keller-Segel model of chemotaxis, introduced in [9] and [11], has provided a cornerstone for much of this work. A scheme recently developed in the finite volume framework (see [1]) treats the discretization of the Keller-Segel model in homogeneous domain where the diffusion tensor is considered to be the identity matrix. In this case, the mesh used for the discretization in space is assumed to satisfy the orthogonality condition (see [2]). In this paper, we propose and analyze the convergence of a generalized finite volume scheme applied to a degenerate chemotaxis model allowing heterogeneities and discontinuities in the diffusion coefficients. This generalization will help to reach more complex applications, especially that chemotaxis phenomenon (attraction or repulsion of cells via
∗The authors would like to thank the National Council for Scientific Research (Lebanon), the Ecole Centrale de Nantes and
the Lebanese University for their support to this work.
1Ecole centrale de Nantes, Laboratoire de Math´ematiques Jean Leray and EDSTIM, CNRS UMR 6629, 1 rue de la No´e, 44321
Nantes, France.
georges.chamoun@ec-nantes.fr, mazen.saad@ec-nantes.fr
2 Lebanese University, Laboratory of Mathematics-EDST and Faculty of Sciences I, Hadath, Liban. rtalhouk@ul.edu.lb
c
EDP Sciences, SMAI 2014
a chemo-attractant or a chemo-repellent) in a heterogeneous medium can yield such discontinuities, since the conductivities of the different components of the medium may be quite different.
2. Setting of the problem
The well-known Keller-Segel model is introduced in general with homogeneous and isotropic diffusion (see [1]). In order to take into account the heterogeneities and the discontinuities in the diffusion coefficients, we consider the following coupled parabolic system:
∂tu − div(s(x)a(u)∇u) + div(s(x)χ(u)∇v) = 0 in QT, (1) ∂tv − div(m(x)∇v) = g(u, v) in QT, (2) with the no-flux boundary conditions on Σt= ∂Ω × (0, T ),
s(x)a(u)∇u · η = 0, m(x)∇v · η = 0, (3) with initial conditions on Ω,
u(x, 0) = u0(x), v(x, 0) = v0(x) . (4) We have QT := (0, T ) × Ω where T > 0 is a fixed time, and let Ω be a bounded domain in Rd, d = 2 or d = 3 where the boundary ∂Ω is Lipschitz and η is the unit outward normal vector.
The species u represents the cell density and v accounts for the chemical concentration. The diffusive flux modelling undirected (random) cell migration and the cross-diffusion flux with velocity dependent on the gradient of the signal, modelling the contribution of chemotaxis. Hillen and Painter were the first who introduced mechanistic descriptions of volume-filling effects (see [4]). Assuming that cells carry a certain finite (nonzero) volume and that occupation of an area limits other cells from penetrating it, a density-dependent chemotactic sensitivity χ(u) function, describing the probability of finding space by a local cell density u, can be derived. It models the migration of species u to location with high concentration of v. The coefficient of diffusion is denoted by a(u). The diffusive functions in a heterogeneous medium are denoted by s(x) and m(x) which may be discontinuous. The function g(u, v) describes the rates of production and degradation of the chemical signal (chemoattractant); here, we assume it is of birth-death structure, i.e., a linear function,
g(u, v) = αu − βv; α, β ≥ 0 . (5) We assume that the density-dependent diffusion coefficient a(u) degenerates for u = 0 and u = um. This means that the diffusion vanishes when u approaches values close to the threshold umand also in the absence of cell-population. Secondly, we assume that χ(0) = 0 and there exists a maximum density of cells um such that χ(um) = 0. The threshold condition has a clear biological interpretation; the cells stop to accumulate at a given point of Ω after their density attains certain threshold value um, therefore the chemotactical sensitiv-ity χ(u) vanishes when u tends to um. This interpretation is called the volume-filling effect, or prevention of overcrowding. The positivity of χ means that the chemical attracts the cells; the repellent case is the one of negative χ.
Upon normalization of um, we can assume that um= 1. Indeed, one can consider the following transforma-tion: ˜u = u
um, ˜v = v, ˜χ(˜u) =
χ(˜uum)
um , ˜a(˜u) = a(˜uum), ˜g(˜u, ˜v) = g(˜uum, ˜v) and we omit the tildas in the notation.
The main assumptions are:
130 ESAIM: PROCEEDINGS AND SURVEYS
χ : [0, 1] 7−→ R is continuous and χ(0) = χ(1) = 0 . (7) A standard example for χ is:
χ(u) = u(1 − u) for u ∈ [0, 1] .
Next, we require general diffusive functions s and m which may be discontinuous such that:
s ∈ L∞(Ω) and there exists ¯s, s ∈ R∗+ such that s ≤ s ≤ ¯s a.e . (8) Definition 2.1. Assume that 0 ≤ u0 ≤ 1, v0 ≥ 0 and v0 ∈ L∞(Ω). A weak solution of (1)-(4) is a pair (u, v) of functions on QT such that
0 ≤ u(x, t) ≤ 1, v(x, t) ≥ 0 a.e. in QT , u ∈ L∞(QT), A(u) := Z u 0 a(r) dr ∈ L2(0, T ; H1(Ω)) , v ∈ L∞(QT) ∩ L2(0, T ; H1(Ω)) , and, for all ϕ and ψ ∈ D([0, T ) × ¯Ω),
− Z Ω u0(x)ϕ(0, x) dx − ZZ QT u∂tϕ dxdt + ZZ QT s(x)∇A(u) · ∇ϕ dxdt − ZZ QT s(x)χ(u)∇v · ∇ϕ dxdt = 0 , − Z Ω v0(x)ψ(0, x) dx − ZZ QT v∂tψ dxdt + ZZ QT m(x)∇v · ∇ψ dxdt = ZZ QT g(u, v)ψ dxdt .
3. Numerical scheme
This section is devoted to the formulation and to the proof of convergence of a finite volume scheme for a Keller-Segel model with general isotropic scalar diffusion functions. We will first describe the space and time discretizations, then we will give the numerical scheme and the result of convergence.
3.1. Space And Time Discretizations.
The space discretization of the domain Ω is based on an admissible mesh as defined in [2] (see Figure 1). It is a finite family Thof polygonal open convex subsets K of Ω, called the control volumes such that ¯Ω = ∪K∈ThK,¯
where h = supK∈T
h(diam(K)), together with a finite family E of disjoint subsets of ¯Ω consisting in non-empty
open convex subsets σ of affine hyperplanes of Rd, called the edges, and a family P = {xK, K ∈ Th} of points in Ω, called the centers verifying the following properties,
• For any edge σ ∈ E, either σ ⊂ ∂Ω or σ = ¯K ∩ ¯L for some K 6= L in Th. In the latter case, we denote σ = σK,L, called the interfaces. We denote by N (K) the control volumes neighbors of K.
• For any K ∈ Th, the point xK belongs to K. For any σK,L∈ E, the line (xK, xL) is orthogonal to σK,L. In addition to that, for any interior edge σK,L, we denote by nK,Land dK,L, respectively, the unit vector normal to σK,Loutward of K and the distance from xK to xL. For any exterior edge σ, the distance is taken from the center xK to the middle point of the edge σ. The measure of K is denoted by |K| and the (d − 1)-dimensional measure of σ ∈ E is denoted by |σ|.
The time discretization is the sequence of discrete times tn = n∆t for n ∈ N, where ∆t > 0 is a given time-step.
Let us consider Thas an admissible mesh, such that the discontinuities of s and m belongs to the interfaces of this mesh. The numerical scheme is obtained by using the finite volume method: equations (1) and (2) of the model are integrated on each control volume K and interval of time (tn, tn+1) and then we shall approximate
Figure 1. A space discretization of Ω .
the normal fluxes over each edge σ of K. Note that if s(x) = 1, the approximation of the normal diffusive flux ∇A(u) · ηK,L on the interface σ = σK,L has been detailed in [1] and the transmissibilities were defined as τK,L = |σdK,L|
K,L . In the following subsections, we will construct an approximation of the normal diffusive flux
s(x)∇A(u) · ηK,L with continuous and discontinuous heterogeneous diffusive functions s(x).
3.2. Continuous Diffusion Coefficients.
In this case, the admissibility assumption on Th allows us to simply approximate this normal diffusive flux by means of the divided differences,
Z σK,L s(x)∇A(u) · ηK,L≈ sK,L |σK,L| dK,L A(uL) − A(uK), (9)
where sK,L denotes the approximation of s(x) on the interface σK,L, with sK,L := s(¯xK,L) such that ¯xK,L is the intersection between the segment [xK, xL] and the common interface σK,L. Consequently, the new transmissiblities in this case are:
τK,L= sK,L |σK,L|
dK,L
. (10)
3.3. Discontinuous Diffusion Coefficients.
In order to treat the case of discontinuities of the diffusion coefficients which lay over the boundaries of the control volumes, let us introduce,
sK = 1 |K| Z K s(x) dx and sK,σ= |sKηK,σ| ,
where |.| is the Euclidean norm, sK,σ be the approximation of s(x) on the edge σ = σK,L and ηK,σ is the unit outward normal vector at σ with respect to K.
In order to obtain the local conservativity, we will introduce auxiliary unknowns uσ on the interfaces. These auxiliary unknowns are helpful to write the scheme, but they can be eliminated locally so that the discrete equations will only be written with respect to the primary unknowns (uK)K∈Th. Since s is continuous on the
132 ESAIM: PROCEEDINGS AND SURVEYS side of σ = σK,Lby using the finite difference principle,
Hσ = sK,σ A(uσ) − A(uK) dK,σ on K; Hσ = sL,σ A(uL) − A(uσ) dL,σ on L ,
where dK,σ (resp. dL,σ) is the distance from xK (resp. xL) to the interface σK,L. Requiring the two above approximations to be equal (the conservativity of the diffusive flux) yields the value of A(uσ),
A(uσ) = 1 sL,σ dL,σ + sK,σ dK,σ A(UL) sL,σ dL,σ + A(UK) sK,σ dK,σ .
This latter allows to give the expression of the approximation Hσ, Hσ= τσ(A(UL) − A(UK)) with τσ=
sK,σsL,σ sL,σdK,σ+ sK,σdL,σ . (11) Consequently, Z σK|L
s(x)∇A(u) · ηK,L≈ τσ|σK,L| A(uL) − A(uK) , (12)
with the following new transmissibilities:
τK,L= τσ|σK,L| . (13) The same guidelines are used to obtain an approximation Gσ of the flux m(x)∇v · ηK,L.
3.4. Numerical Scheme.
In addition to the previous subsections, we still have to approximate s(x)χ(u)∇v ·ηK,Lby means of the values UK, ULand δVK,L(where δVK,Ldenotes the approximation of the flux s(x)∇v · ηK,Lon σK,L) that are available in the neighborhood of the interface σK,L. To do this, we use a numerical flux function G(UK, UL, δVK,L). Numerical convection flux functions G of arguments (a, b, c) ∈ R3, are required as in [1] to satisfy the following properties,
• G(., b, c) is non-decreasing for all b, c ∈ R, and G(a, ., c) is non-increasing for all a, c ∈ R ; • G(a, b, c) = −G(b, a, −c) for all a, b, c ∈ R; hence the flux is conservative.
• G(a, a, c) = χ(a)c for all a, c ∈ R ; hence the flux is consistent.
• There exists C > 0, such that for all a, b, c ∈ R, |G(a, b, c)| ≤ C(|a| + |b|)|c|. • |G(a, b, c) − G(a0, b0, c)| ≤ |c|(|a − a0| + |b − b0|) for all a, b, a0, b0, c ∈ R.
Remark 3.1. One possibility to construct the numerical flux G is to split χ in the non-decreasing part χ↑ and the non-increasing part χ↓, such that
χ↑(z) := Z z 0 (χ0(s))+ds, χ↓(z) := Z z 0 (χ0(s))− ds . Herein s+= max(s, 0) and s− = max(−s, 0). Then we take,
G(a, b, c) = c+(χ↑(a) + χ↓(b)) − c−(χ↑(b) + χ↓(a)) . (14) Notice that in the case χ has unique local (and global) maximum at the point ¯u ∈ [0, 1], such as the flux χ(u) = u(1 − u), we have
Figure 2. An admissible triangulations Th: Mesh of 1253 triangles with an empty hall (left) and of 3584 triangles (right) .
Finally, we obtain the following scheme: ∀K ∈ Th, UK0 = 1 |K| Z K u0(x) dx, VK0 = 1 |K| Z K v0(x) dx , (15) and ∀n ∈ [0, ..., N ], |K|U n+1 K − U n K ∆t − X L∈N (K) τK|L A(ULn+1) − A(U n+1 K ) + X L∈N (K) G(UKn+1, ULn+1; δVK,Ln+1) = 0 , (16) |K|V n+1 K − V n K ∆t − X L∈N (K) µK,L VLn+1− V n+1 K = |K|g(U n K, V n+1 K ) , (17) where δVK,Ln+1= τK|L(VLn+1− V n+1
K ), τK,L (resp. µK,L) are the new transmissibilities defined in (10) (or (13)) and the unknowns are U = (UKn+1)K∈Th and V = (V
n+1
K )K∈Th, n ∈ [0..N ]. The discrete solution associated to
this discrete problem is, (uh, vh) defined as constant functions on QT given by, ∀K ∈ Th, ∀n ∈ [0, ..., N ], uh ]tn,tn+1]×K= U n+1 K , vh ]tn,tn+1]×K = V n+1 K . For this numerical scheme, we have proved the following result.
Theorem 3.1. Assume (5), (7), (6) and (8). Consider v0∈ L∞(Ω), v0≥ 0 and 0 ≤ u0≤ 1 a.e. on Ω. 1) There exists a solution (uh, vh) of the discrete system (16)-(17) with initial data (15).
2) Any sequence (hm)m decreasing to zero possesses a subsequence, still denoted as the sequence, such that (uhm, vhm) converges a.e. on QT to a weak solution (u, v) of the modified Keller-Segel system (1)-(4) in the
sense of Definition 2.1 .
Outline of the Proof. The discrete maximum principle still verified due to the positivity of the new transmissibilities defined in (10) and in (13). Consequently, one can maintain the same necessary estimates of [1] to prove the existence of a discrete solution. Then, estimates in time and space are constructed to use the Kolmogorov compactness criterion and to conclude the existence of a subsequence (uh, vh) of discrete solutions that converges to a function (u, v) almost everywhere in [0, T ] × Ω. Finally, we proved that this function (u, v) is a weak solution of the modified Keller-Segel model.
134 ESAIM: PROCEEDINGS AND SURVEYS
Figure 3. Test 1- Initial conditions for the cell density u0 (left) and for the concentration of the chemo-attractant v0 (right) .
Figure 4. Test 1- Evolution of the cell density (u), at time t = 5 with 0 ≤ u ≤ 0.3283 (left), at time t = 12.5 with 0 ≤ u ≤ 0.2413 and at time t = 17.5 with 0 ≤ u ≤ 0.2150 (right) .
4. Numerical tests
In this section, we shall illustrate the behavior of the discrete solutions of the proposed numerical scheme for isotropic heterogeneous and discontinuous coefficients. The computations were done with numerical handwork Fortran 95 code, where this scheme is implemented. The algorithm used to compute numerical solution of the discrete problem is the following: at each time step, we first calculate Vn+1solution of the linear system given by the equation of (17) and next we compute Un+1as the solution of the nonlinear system defined by the first equation of (16). For this end, a Newton algorithm is implemented to approach the solution of nonlinear system and a bigradient method to solve linear systems arising from the Newton algorithm process. We will provide our tests on admissible meshes given in the Figure 2.
Figure 5. Test 1- The cell density (u) at time t = 50 with 0 ≤ u ≤ 0.4973 (left), at time t = 100 with 0 ≤ u ≤ 0.7047 (right) .
Figure 6. Test 2- Initial conditions for the cell density u0 (left) and for the concentration of the chemo-attractant v0 (right) .
Test 1: Continuous heterogeneous case. Let us consider the following data: Lx = 3, Ly = 3 as the length and the width of the domain given in the Figure 1 (left). In this first test, we consider the diffusion functions as
s(x) = (x − 1.5)2+ (y − 1.5)2; m(x) = 1. Further, dt = 0.005, α = 0.01, β = 5 × 10−5, A(u) = D(u22 − u3
3), with D = 10
−1, χ(u) = cu(1 − u)2, with c = 10−1. Finally, the diffusion coefficient of the chemo-attractant is d = 10−4. The initial conditions are defined by regions. The initial density is defined as u0(x, y) = 1 in the square (x, y) ∈ [0.2, 0.8] × [1.2, 1.8]
and 0 otherwise. The initial chemoattractant is defined as v0(x, y) = 5 in the union of two squares (x, y) ∈ [1.2, 1.8] × [0.2, 0.8] ∪ [1.2, 1.8] × [2.2, 2.8] and 0 otherwise (see Figure 3). In Figures 4 and 5, we show the evolution of the cell density. We observe during the stage of evolution the effect of the chemo-attractant, since the cells are present in the chemoattractant regions.
136 ESAIM: PROCEEDINGS AND SURVEYS
Figure 7. Test 2- The cell density (u) at time t = 2.5 with 0 ≤ u ≤ 0.1055 (left) and at time t = 5 with 0 ≤ u ≤ 0.07963 (right) .
Figure 8. Test 2- The cell density (u) at time t = 10 with 0 ≤ u ≤ 0.1991 (left) and at time t = 40 with 0 ≤ u ≤ 0.5570 (right) .
Test 2: Discontinuous case. In this test, our target is to prove the efficiency of our numerical scheme in treating discontinuous diffusion. For that, let us consider
s(x) =
6 if y ≤ 0.5
1 if y > 0.5 , m(x) = 1 .
The space domain Ω is the unit square and the mesh is given in the Figure 1 (right). One can remark that the discontinuities of the diffusion coefficients coincide with the interfaces of the mesh. Further, dt = 0.005, α = 0.01, β = 0.05, A(u) = D(u22−u3
3) with D = 0.03, χ(u) = cu(1 − u)
2with c = 0.1. Finally, the diffusion coefficient of the chemo-attractant is d = 10−5. The initial conditions are also defined by regions. The initial density is defined as u0(x, y) = 1 in the square (x, y) ∈ [0.15, 0.25] × [0.45, 0.55] and 0 otherwise. The initial chemoattractant is defined as v0(x, y) = 10 in the union of two squares (x, y) ∈ [0.45, 0.55]×[0.7, 0.8] ∪ [0.45, 0.55]×[0.2, 0.3] and 0 otherwise (see Figure 6). The behavior of the cell density via the chemo-attractant and the influence of the
discontinuous diffusion coefficients are clear, as can be seen in the Figures 7 and 8. We observe during the stage of evolution the effect of the anisotropic diffusion since the cells are more present in the down chemoattractant region.
5. conclusion
In this article, we propose a variant of the Keller-Segel model and a finite volume numerical method to simulate this chemotaxis model with general scalar diffusion functions. The approximate solutions remains biologically admissible due to the confinement of the cell density as a consequence of the discrete maximum principle. The convergence to a weak solution of the continuous model is guaranteed and the numerical exper-iments allow the validation of the generalized numerical scheme.
References
[1] B. Andreianov, M. Bendahmane and M. Saad, Finite volume methods for degenerate chemotaxis model. Journal of computational and applied mathematics, 235: p. 4015-4031, 2011.
[2] R. Eymard, T. Gallouet and R. Herbin, Finite volume methods, Handbook of numerical analysis. Handbook of numerical analysis, Vol VII North-Holland, Amsterdam, p. 713-1020, 2000.
[3] K. Brenner, M´ethode de volumes finis sur maillages quelconques pour des syst`emes d’´evolution non lin´eaires. Th`ese soutenue `
a l’Universit´e Paris XI, le 8 novembre 2011.
[4] T. Hillen and K. Painter, A user’s guide to PDE models for chemotaxis. J Math Biol.; 58(1-2), p. 183-217, 2009.
[5] R. Herbin and F. Hubert, Benchmark on discretization schemes for anisotropic diffusion problems on general grids. 5th International Swnposium on Finite Volumes For Complex Applications, 2008.
[6] R. Eymard, T. Gallouet, R. Herbin and J.C. Latche, Analysis tools for finite volume schemes. Acta Math. Univ. Comenianae, Vol. LXXVI, 1, p. 111-136, 2007.
[7] J.L. Lions, Quelques m´ethodes de r´esolution des probl`emes aux limites non-lin´eaires. Dunod, Paris, 1969. [8] B. Dibenedetto, Degenerate Parabolic Equations. Springer-Verlag, New York, 1993.
[9] D. Horstmann, from 1970 until present, The keller-Segel model in chemotaxis and its consequences. I.Jahresberichte DMV 105 (3), p. 103-165, 2003.
[10] R. Eymard, T. Gallouet, R. Herbin and A. Michel, Convergence of a finite volume scheme for nonlinear degenerate parabolic equations. Numer .Math. 92(1), p. 41-82, 2002.
[11] E.F. Keller and L.A. Segel, The Keller-Segel model of chemotaxis. J Theor Biol. 26, p. 399-415, 1970. [12] H. Brezis, Analyse fonctionnelle, Th´eorie et Applications, Masson, Paris, 1983.
